Put your coding agent on a schedule
Cron jobs and a template gallery for the recurring work: the nightly build, the dependency bump, the stale-issue sweep. They run on your own machine, under the same policy as everything else. No run cap from us; your model plan's limits, and your vendor's terms for unattended use, still apply and may change.
Start from a template
The dashboard's Schedule page opens a gallery of 10 ready-made jobs for coding work. Picking one fills in the form; you review the prompt and the cadence and save it before anything runs. The ones marked Writes to your repos push branches, open pull requests or change issues, so they sit behind the same approval gate as any other tool call.
Code quality
-
Nightly Build Watch
Builds and tests main overnight; reports failures and likely fixes.
Daily · 02:00
-
CI Failure Triage
Reproduces newly failed checks and opens a fix PR with test evidence.
Every 30 min Writes to your repos
-
Review Follow-Through
Summarizes new review comments and what each would take to address.
Every 30 min
-
Bug Intake & Repro
Turns new bug issues into a reproduction and a failing test.
Every 30 min Writes to your repos
-
Merged-PR Checklist
Grades yesterday's merges against the team checklist, read-only.
Daily · 08:30
-
Coverage Backfill
Writes real tests for the most under-covered recently changed files.
Weekly · Wednesday 09:00 Writes to your repos
-
Lint & Type-Check Sweep
Fixes lint and type regressions on the default branch, no suppressions.
Daily · 07:00 Writes to your repos
Repo hygiene
-
Dependency Guardian
Upgrades packages, runs tests, and opens a PR only when green.
Weekly · Monday 06:00 Writes to your repos
-
Issue Labeler & Stale Sweep
Labels new issues and runs a courteous two-step stale process.
Daily · 08:00 Writes to your repos
-
Docs Drift Detector
Diffs merged changes against docs and opens fix PRs for drift.
Weekly · Tuesday 10:00 Writes to your repos
Template names, descriptions and default cadences as they ship in the dashboard today. Every cadence is editable before you save.
The week, on one page
The calendar view lays each job across the week, so a 02:00 build and a 06:00 dependency bump do not land on the same hour by accident. The executions view keeps each run's result.
No night has been recorded for this page yet: the plates show the dashboard with seeded jobs and no agent connected, and they say so. A recorded night will carry its harness and date in the caption.
Plate I. The template gallery
Capture pending: this plate will hold a real screenshot with its harness and date in the frame.
Plate II. The week's schedule
Capture pending: this plate will hold a real screenshot with its harness and date in the frame.
Seeded jobs, not runs.
Where you can create a job, on which harness
A job is created by one of three surfaces. Two of them work with any harness you dock. The third, asking in chat, needs the agent to call its own scheduling tool, and those tools reach kiro-cli only today.
| Surface | Works with | How |
|---|---|---|
| Dashboard | every docked harness | Schedule page: a template, or the form (name, prompt, interval or cron expression, optional agent) |
| CLI | every docked harness | warding cron add, warding cron list, warding cron remove |
| Chat: create a job | kiro-cli only, today | "check the build every night at two"; the agent calls its scheduling tool |
| Chat: manage jobs | every docked harness | /cron list | pause | resume | remove on Telegram; cron list and friends on Slack |
warding cron add "nightly-build" "build and test main; report failures" --cron "0 2 * * *"
warding cron add "ci-watch" "triage new CI failures" --every 1800
warding cron list
warding cron remove <id>
What a scheduled run is
- Three kinds of schedule. An interval (at least 60 seconds), a five-field cron expression, or a one-shot at a set time.
- A fresh agent session per job, reused across runs by default so a digest can refer to last night; set it to start clean each run for scanners and polling.
- Under the same gate. A run's tool calls go through the approval ladder and the deny rules like any other session. A job can be set to auto-approve; a deny rule still refuses.
- Results go where the job came from: the dashboard bell, and the chat conversation a job was created in. A failed run delivers its reason, not only a log line; repeated failures pause the job.
- Skip dates and a time zone per job, for holidays. Daily and hourly jobs get a small random start delay; mark a job strict to fire on the minute.
- A time limit per run: 30 minutes by default, up to 24 hours per job. Longer, checkpointed work belongs in the task runner.
Limits, stated plainly
- No run cap from us; your model plan's limits still apply. We add no daily count. Your vendor's usage limits and its terms for unattended use are theirs, and they can change.
- The box has to be on. A job fires only while the gateway is running. Nothing of ours runs in the cloud.
- One harness for jobs. Every job runs on the harness docked for chat when it fires; switching is one setting.
- Unverified overnight. No scheduled night has been recorded on any harness yet. The verification log gets the dated row when one is. Being verified
Already on Claude Code? Anthropic has Routines and Desktop scheduled tasks, with no setup. When to use those, and when to use this.
Review the local beta.
Apache-2.0 source publication is pending. Read the setup notes and verification limits.
Source publication pending
Public source, install commands and downloads are not available yet. A clean install and live workflow are not yet verified.